Back to Models

GPT Image 2

openaiopenai/gpt-image-2

OpenAI's most advanced image generation model with native reasoning β€” thinks before drawing. 2K resolution, multi-image consistency, magazine-quality typography, and image editing. Released April 21, 2026.

GPT Image 2 (gpt-image-2), launched April 21, 2026 under the "ChatGPT Images 2.0" brand, is OpenAI's most capable image generation model. It is the first OpenAI image model to integrate O-series reasoning capabilities β€” planning the composition, searching the web for references, and self-verifying outputs before rendering a single pixel. It represents a step change from the diffusion-based pipelines of DALLΒ·E 3 and gpt-image-1: the model reasons over the brief first, then generates, making it an Agentic image model rather than a simple text-to-image decoder.

The model delivers native 2048Γ—2048 (2K) resolution output, best-in-class multilingual text rendering across CJK, Cyrillic, Arabic, Hindi, and Bengali scripts, up to 8 consistent images per prompt with shared characters and lighting, transparent background support, and functional QR code generation. On Image Arena, GPT Image 2 took #1 in every category at launch with a +242 point lead in Text-to-Image β€” the largest margin ever recorded on the platform. Through TheRouter, the model is accessible via the standard OpenAI-compatible /v1/images/generations endpoint, making it a drop-in upgrade from gpt-image-1.5 or DALLΒ·E 3.

Best for
  • β€’ Text-heavy marketing collateral β€” posters, infographics, ads, and UI mockups where text inside the image must be sharp, accurate, and correctly rendered in CJK, Cyrillic, Arabic, or mixed scripts
  • β€’ Thinking-mode complex layouts β€” multi-page infographics, manga panels, storyboards, scientific diagrams, and editorial layouts where the model's reasoning pass ensures spatial accuracy
  • β€’ Multi-image consistent assets β€” up to 8 images per prompt with shared characters, lighting, color palette, and style, ideal for campaign creative suites and product variant shots
  • β€’ Localized creative production β€” native-language posters, ads, and packaging for international markets with character-level accuracy in non-Latin scripts previously unattainable in AI image generation
  • β€’ Functional QR code generation β€” QR codes that actually scan, styled with brand colors and embedded in designed posters, collapsing three tools into one API call
Reach for something else if
  • β€’ Ultra-high-volume batch generation on low-tier accounts β€” Tier 1 caps at 5 images/minute; reaching Tier 3 (50/min) requires $100 spend and a 7-day-old account. Use gpt-image-1-mini (~75% cheaper) for drafts
  • β€’ Latency-sensitive real-time pipelines β€” thinking mode can take up to 2 minutes for complex prompts; instant mode is faster but still lacks streaming support (no incremental output)
  • β€’ Open-source or offline deployment β€” gpt-image-2 is proprietary and available only via API; no self-hosting or fine-tuning option exists at launch
  • β€’ Highly stylized painterly/illustrative work β€” Midjourney V8 retains an edge for editorial illustration, book covers, and artistic aesthetics where mood matters more than precision

How TheRouter serves this differently from the vendor

As the vendor operates it

OpenAI serves direct gpt-image-2 calls on /v1/images/generations and /v1/images/edits; its model page marks /v1/responses and /v1/chat/completions as not supported for the model ID itself.

On TheRouter

TheRouter exposes the same image generation and image edit surfaces under https://api.therouter.ai/v1 with the routed model id openai/gpt-image-2. Examples therefore keep the OpenAI SDK shape but change baseURL, API key, and model slug.

Context Length
--
Max Output
--
Input Priceper 1M tokens
$5.40/ 1M tokens
Output Priceper 1M tokens
$32.40/ 1M tokens

Modalities

textimage→image

Capabilities

VisionImage GenerationImage Edit

Media Generation Capabilities

image_generation
sizes
  • 1024x1024
  • 1024x1536
  • 1536x1024
qualities
  • low
  • medium
  • high
max_outputs_per_request
4
output_formats
  • png
  • jpeg
  • webp
defaults
size
1024x1024
quality
medium
image_edit
sizes
  • 1024x1024
  • 1024x1536
  • 1536x1024
qualities
  • low
  • medium
  • high
max_reference_images
4
output_formats
  • png
  • jpeg
  • webp
defaults
size
1024x1024
quality
medium

Pricing Breakdown

TypeRate
Input$5.40 / 1M tokens
Output$32.40 / 1M tokens
Image input$8.64 / 1M tokens

Billed per-token (input text, image-input tokens, image-output tokens) β€” see /v1/models pricing fields, not a flat per-image price.

Supported Parameters

promptsizequalitybackgroundoutput_formatnuser

Specifications

Release date2026-04-21openai.com β†—verified
Max resolutionUp to 3840px on the long edge; popular 4K sizes include 3840Γ—2160 and 2160Γ—3840developers.openai.com β†—verified
Aspect ratiosContinuous sizes where the long-edge:short-edge ratio does not exceed 3:1; both edges must be multiples of 16pxdevelopers.openai.com β†—verified
Output formatsPNG by default; JPEG and WebP are also supported. Transparent backgrounds are not supported for gpt-image-2.developers.openai.com β†—verified
Knowledge cutoffNot publicly discloseddevelopers.openai.com β†—unknown
Thinking modeOptional reasoning pass before generation β€” web search, layout planning, self-verification (off by default in API)openai.com β†—verified
Images per promptMultiple images in a single request via the n parameter; exact maximum not publicly disclosed in the model guidedevelopers.openai.com β†—unknown
LicenseProprietary (OpenAI API terms β€” no training on customer data)verified

Benchmarks

BenchmarkDistributionScoreSource
Image Arena β€” Text-to-Image (overall)
#1 β€” +242 lead over Nano-banana-2, largest margin in Arena history
1512arena.ai β†—
Image Arena β€” Single-Image Edit
#1 β€” +125 over Nano-banana-pro
1513arena.ai β†—
Image Arena β€” Multi-Image Edit
#1 β€” +90 over Nano-banana-2
1464arena.ai β†—
Image Arena β€” Text Rendering sub-category
Largest improvement over gpt-image-1.5 β€” +316 point jump in text rendering quality
+316arena.ai β†—
Image Arena β€” Product, Branding & Commercial Design
+277 over gpt-image-1.5
+277arena.ai β†—
Image Arena β€” Portraits sub-category
+296 over gpt-image-1.5
+296arena.ai β†—

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

Recommended: use the async API

Image generation typically takes 30–180s, beyond the edge sync timeout. The examples below use the ?async=true submit + poll pattern. Read the full async image generation & edit guide β†’

cURL
# 1) Submit job (returns 202 immediately with a polling URL).
# Image generation takes 30-180s β€” always use the async path in production.
JOB=$(curl -s -X POST "https://api.therouter.ai/v1/images/generations?async=true"   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "openai/gpt-image-2",
    "prompt": "A cinematic product render with soft studio lighting"
  }' | python3 -c "import sys,json;print(json.load(sys.stdin)['id'])")
echo "submitted: $JOB"

# 2) Poll until terminal (succeeded / failed / cancelled / expired).
while :; do
  R=$(curl -s "https://api.therouter.ai/v1/jobs/$JOB"     -H "Authorization: Bearer $THE_ROUTER_API_KEY")
  S=$(echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['status'])")
  echo "status: $S"
  case "$S" in
    succeeded) echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['unsigned_urls'][0])"; break ;;
    failed|cancelled|expired) echo "$R"; exit 1 ;;
  esac
  sleep 5
done

Image generation

Generate images from text prompts using the standard OpenAI Images API. Supports 2K resolution, quality tiers, output format selection, transparent backgrounds, and up to 8 images per batch.

cURL
curl https://api.therouter.ai/v1/images/generations \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2",
    "prompt": "A serene mountain landscape at sunset, photorealistic style",
    "n": 1,
    "size": "2048x2048",
    "quality": "medium",
    "output_format": "png"
  }'

Image Editing Examples

Upload an image and describe the edit you want with a text prompt; the model returns the edited image as base64.

cURL
# Same async submit + poll pattern as /v1/images/generations.
JOB=$(curl -s -X POST "https://api.therouter.ai/v1/images/edits?async=true"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -F "model=openai/gpt-image-2"   -F "prompt=Turn this scene into a watercolor painting"   -F "size=1024x1024"   -F "image=@input.png" | python3 -c "import sys,json;print(json.load(sys.stdin)['id'])")
echo "submitted: $JOB"

while :; do
  R=$(curl -s "https://api.therouter.ai/v1/jobs/$JOB"     -H "Authorization: Bearer $THE_ROUTER_API_KEY")
  S=$(echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['status'])")
  echo "status: $S"
  case "$S" in
    succeeded) echo "$R" | python3 -c "import sys,json;print(json.load(sys.stdin)['unsigned_urls'][0])"; break ;;
    failed|cancelled|expired) echo "$R"; exit 1 ;;
  esac
  sleep 5
done

More from openai

Similar models

Cross-provider sibling models

News & changes

2026-04-21

OpenAI launches ChatGPT Images 2.0 with gpt-image-2 β€” reasoning-powered image generation

OpenAI unveiled gpt-image-2, its most capable image generation model, with native 2K resolution, O-series reasoning capabilities, best-in-class multilingual text rendering across CJK, Arabic, Hindi, and Bengali scripts, up to 8 consistent images per prompt, web search integration, and functional QR code generation. The model took #1 in every Image Arena category on launch day with record-breaking margins. DALLΒ·E 2 and DALLΒ·E 3 set for retirement on May 12, 2026, making gpt-image-2 the default across OpenAI's platform.

re-authored by TheRouteropenai.com β†—

Related news

Frequently asked

How is gpt-image-2 different from gpt-image-1.5?

gpt-image-2 is a major architectural upgrade β€” the first OpenAI image model with O-series reasoning capabilities. It introduces thinking mode (the model plans the composition, searches the web, and self-verifies before generating), native 2K (2048Γ—2048) resolution, near-perfect multilingual text rendering across CJK, Arabic, Hindi, Bengali, and Cyrillic scripts, up to 8 consistent images per prompt, web search integration, and functional QR code generation. On benchmarks, it achieved +242 point lead on Image Arena Text-to-Image vs. the next best model. DALLΒ·E 3 is set to retire on May 12, 2026, with gpt-image-2 becoming the default.

What resolution and aspect ratios does gpt-image-2 support?

gpt-image-2 supports native output resolution up to 2048Γ—2048 (2K), with a continuous aspect ratio range from 3:1 (wide banner) to 1:3 (tall vertical). This eliminates the need for cropping or upscaling after generation. The previous generation was limited to fixed presets like 1:1, 2:3, and 3:2. Lower resolutions can be specified for faster generation and lower costs.

re-authored by TheRouterfelloai.com β†—
Is gpt-image-2 compatible with the existing DALLΒ·E 3 integration I already have?

Yes β€” the migration is mostly a model-ID swap. The API uses the same /v1/images/generations and /v1/images/edits endpoints. Change your model parameter from "dall-e-3" to "openai/gpt-image-2" (or "gpt-image-2" if calling OpenAI directly). DALLΒ·E 2 and DALLΒ·E 3 are officially retiring on May 12, 2026, so migration is required. TheRouter users simply update the model slug in their API call.

re-authored by TheRoutergptimage2.design β†—
Fact ledger β€” every claim on this page traces here
sourceURLretrieved
Release dateopenai.com β†—2026-05-29verified
Max resolutiondevelopers.openai.com β†—2026-08-04verified
Aspect ratiosdevelopers.openai.com β†—2026-08-04verified
Output formatsdevelopers.openai.com β†—2026-08-04verified
Knowledge cutoffdevelopers.openai.com β†—2026-08-04unknown
Thinking modeopenai.com β†—2026-05-29verified
Images per promptdevelopers.openai.com β†—2026-08-04unknown
Licenseβ€”β€”verified
Image Arena β€” Text-to-Image (overall)arena.ai β†—2026-08-04verified
Image Arena β€” Single-Image Editarena.ai β†—2026-08-04verified
Image Arena β€” Multi-Image Editarena.ai β†—2026-08-04verified
Image Arena β€” Text Rendering sub-categoryarena.ai β†—2026-08-04verified
Image Arena β€” Product, Branding & Commercial Designarena.ai β†—2026-08-04verified
Image Arena β€” Portraits sub-categoryarena.ai β†—2026-08-04verified
OpenAI launches ChatGPT Images 2.0 with gpt-image-2 β€” reasoning-powered image generationopenai.com β†—2026-05-29verified
How is gpt-image-2 different from gpt-image-1.5?openai.com β†—2026-05-29to verify
What resolution and aspect ratios does gpt-image-2 support?felloai.com β†—2026-05-29to verify
Is gpt-image-2 compatible with the existing DALLΒ·E 3 integration I already have?gptimage2.design β†—2026-05-29to verify
Help & contact